SaLinA: Sequential Learning of Agents

TL;DR

SaLinA is a user-friendly, flexible library built on PyTorch that simplifies implementing complex sequential learning models, including various reinforcement learning algorithms, and supports large-scale training across multiple CPUs and GPUs.

Contribution

It introduces SaLinA, a versatile library that lowers the barrier to implementing diverse sequential learning models and integrates seamlessly with existing deep learning workflows.

Findings

Supports multiple RL paradigms like model-based, batch, hierarchical, and multi-agent RL.

Enables easy understanding and modification for PyTorch users.

Efficient large-scale training with multi-CPU and GPU support.

Abstract

SaLinA is a simple library that makes implementing complex sequential learning models easy, including reinforcement learning algorithms. It is built as an extension of PyTorch: algorithms coded with \SALINA{} can be understood in few minutes by PyTorch users and modified easily. Moreover, SaLinA naturally works with multiple CPUs and GPUs at train and test time, thus being a good fit for the large-scale training use cases. In comparison to existing RL libraries, SaLinA has a very low adoption cost and capture a large variety of settings (model-based RL, batch RL, hierarchical RL, multi-agent RL, etc.). But SaLinA does not only target RL practitioners, it aims at providing sequential learning capabilities to any deep learning programmer.